⚔️ Optimize truncate_with_protected_tokens to use vectorized operations #3875
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Replace list comprehension with torch.isin for protected token checking. This reduces CPU-GPU synchronization from O(seq_len) to O(1) per sequence, achieving ~9x speedup for typical batch sizes.
Contributor
There was a problem hiding this comment.
Pull Request Overview
This PR optimizes the truncate_with_protected_tokens function in the GRPO trainer by replacing inefficient list comprehensions with vectorized tensor operations. The optimization significantly reduces CPU-GPU synchronization overhead and improves performance.
Key changes:
- Replaces list comprehension with
torch.isin()for vectorized membership testing - Eliminates O(seq_len) CPU-GPU synchronization points per sequence
- Achieves ~9x speedup on typical workloads
kashif
approved these changes
Aug 11, 2025
|
The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update. |
kashif
reviewed
Aug 12, 2025
qgallouedec
changed the title
qgallouedec
approved these changes
Aug 17, 2025
LuisVasquezBSC
pushed a commit
to langtech-bsc/trl
that referenced
this pull request
Aug 28, 2025
…ns (huggingface#3875) Co-authored-by: Kashif Rasul <[email protected]> Co-authored-by: Quentin Gallouédec <[email protected]>
LuisVasquezBSC
pushed a commit
to langtech-bsc/trl
that referenced
this pull request
Aug 28, 2025
…ns (huggingface#3875) Co-authored-by: Kashif Rasul <[email protected]> Co-authored-by: Quentin Gallouédec <[email protected]>
SamY724
pushed a commit
to SamY724/trl
that referenced
this pull request
Sep 6, 2025
…ns (huggingface#3875) Co-authored-by: Kashif Rasul <[email protected]> Co-authored-by: Quentin Gallouédec <[email protected]>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
What does this PR do?
This PR optimizes the
truncate_with_protected_tokensfunction in GRPO trainer by replacing list comprehensions with vectorized tensor operations. The optimization reduces CPU-GPU synchronization overhead from O(seq_len) to O(1) per sequence.Performance improvement
The original implementation used a list comprehension that called
.item()for every token in the sequence to check if it's in the protected set:The optimized version uses
torch.isinfor vectorized membership testing:Benchmark results
Before submitting
Pull Request section?
to it if that's the case.
Who can review?
Anyone in the community is free to review the PR once the tests have passed. Feel free to tag
members/contributors who may be interested in your PR.